@younndai/lyt-llm
The LLM gateway — multi-provider routing for AI-assisted vault operations, with hard cost budgets, on your terms.
Part of the Lyt toolchain — federated markdown vaults you own, made legible to AI.
Website · Repository · Apache 2.0 · Trademark Policy · Contributing
Public alpha — under active testing. Lyt works and we use it daily, but surfaces change between releases and docs are still growing. Install only via the
alphadist-tag. Your vaults are plain markdown in plain git repos — your data is never locked in, and Lyt never phones home.
What is this?
@younndai/lyt-llm is the LLM gateway of Lyt — multi-source LLM routing for AI-assisted vault operations, designed around one principle: you choose where intelligence runs. Local model via Ollama, your own provider keys, your existing Claude Code / Codex subscription, or external providers through a relay — the gateway composes four adapters behind a single createLlmGateway() surface, and your knowledge only goes where you point it.
This keeps Lyt's AI-first posture honest on the model side too: automations that need an LLM step declare a source preference, not a hard-wired vendor, and a per-run cost-budget hard-stop guarantees a runaway automation can't burn through an API budget. No hidden defaults, no silent cloud calls — the gateway is configuration you can read.
You usually do not install this package directly — it is pulled in by @younndai/lyt and consumed by @younndai/lyt-runner.
Install (standalone)
npm install -g @younndai/lyt-llm@alphaThe four adapters
| Adapter | Routes to | Use when |
|---|---|---|
| harness | Claude Code / Codex skill invocation | You already pay for an agent subscription — reuse it, no extra keys |
| ollama | OpenAI-compatible local HTTP at localhost:11434 |
Fully local inference — knowledge never leaves the machine |
| byok | Your own provider keys (Anthropic, OpenAI, …) | Direct provider access under your own account + billing |
| ai-relay | External providers via @younndai/ai-relay |
Managed multi-provider routing |
Routing honours per-automator source_preference + hard_constraints — for example "local only, never external" is a declared, enforced constraint, not a hope.
Cost control
- Per-run hard-stop — a run that exceeds its declared budget is stopped, not billed onward.
- Monthly budget — warn-level in v1.
Programmatic use
import { createLlmGateway } from "@younndai/lyt-llm";The gateway registers std:llm.generate@v1, std:llm.stream@v1, std:llm.generate_object@v1, and std:llm.embed@v1 ops with @younndai/lyt-runner, so YON automators can declare LLM steps like any other governed operation.
The Lyt toolchain
| Package | Role |
|---|---|
@younndai/lyt |
The unified lyt CLI (meta package) |
@younndai/lyt-vault |
The vault primitive |
@younndai/lyt-mesh |
The mesh layer — multi-vault operations |
@younndai/lyt-skills |
Agent-harness skills (Claude Code, Codex, agents) |
@younndai/lyt-mcp |
The MCP server |
@younndai/lyt-runner |
The YON automation runner |
@younndai/lyt-llm |
This package — the LLM gateway |
About YounndAI
YounndAI — You and AI, unified. (pronounced "yoon-dye")
A philosophy of intelligence: building with intention, so humans and machines think together without losing what makes either whole.
License & Attribution
Apache-2.0. 2026 MARLINK TRADING SRL (YounndAI). See LICENSE and NOTICE.
"LYT", "Lyt", and "YounndAI" are trademarks of MARLINK TRADING SRL — see TRADEMARK.md.
Created by Alexandru Mareș.
Website: linkyourthink.com
Structure before scale. Harmony above all.
| Project | Lyt — Link Your Think |
| Author | Alexandru Mareș |
| Company | MARLINK TRADING SRL · YounndAI |
| License | Apache 2.0 — 2026 MARLINK TRADING SRL |
| Trademark | YounndAI Trademark Guidelines |